For a long time, the only way to get a piece of software was to buy it. The firm paid once, the firm owned a copy, and the firm ran it on its own machines for as long as the firm wanted. Then for a long time, the only way to get a piece of software was to rent it. The firm paid every month, the firm did not own a copy, and the firm ran whatever the vendor was running that quarter.
Subscriptions made sense for software that genuinely needed to be operated by someone else. Email infrastructure. Payment processing. Anything that required round-the-clock uptime and a security team larger than the firm itself. The math worked because the alternative, running it in-house, was actually harder.
AI is not that kind of software.
The price math
A small firm pays for an AI subscription the same way it pays for any other subscription: per seat, per month, with a price that goes up every year. Across a fifteen-attorney firm, a fifty-dollar-per-seat-per-month tool is nine thousand dollars a year. A two-hundred-dollar tool is thirty-six thousand. Over five years, a small firm with two such tools is paying a quarter of a million dollars to rent software that, in the meantime, has not been built for the firm in particular.
A bespoke build, scoped to the firm's actual workflow, runs once and stays. The firm pays for the engagement, owns the code, and runs it on its own model access. The recurring cost is the per-call cost of the model itself, which is dropping every year, not rising.
The data math
An AI tool that sits between the firm and the firm's own files is, by construction, a tool that has read the firm's files. Where those files go after they are read, what the vendor is allowed to train on, and how the firm gets out cleanly if the vendor changes its terms are all things that a subscription contract is allowed to leave vague.
A firm-owned system runs in the firm's environment, against an account the firm owns. The data stays in one place. The exit strategy is "we keep using what we built."
The practice math
The hardest argument against subscriptions is not financial. It is that a generic tool, by definition, was not built for any one firm. It cannot read the firm's prior memos in the firm's voice. It cannot draft the way the senior associate would draft. It does not know which intake forms the firm actually uses, which clients are sensitive to which language, or which positions the firm has taken before on which issues.
It can be configured to do some of those things, partially, by the firm spending its own time configuring. But the firm is then doing the customization work for free, on top of paying the subscription, with no equity in the result. When the vendor pivots or shuts down, the customization goes with it.
What we do
Maule & Co. designs and builds the version of the tool the firm would have built if the firm had the time. The engagement runs six to twelve weeks. The firm owns the code at the end. There is no per-seat fee, no monthly invoice, and no third party between the firm and its data. When the underlying technology has advanced enough that the system should be rebuilt rather than patched, we come back for that engagement. Until then, the firm runs what we built.
It is the older deal. The firm pays once. The firm owns a copy. The firm runs it for as long as the firm wants.